Opinion-Polarity Identification in Bengali

In this paper, opinion polarity classification on news texts has been carried out for a less privileged language Bengali using Support Vector Machine (SVM) 1 . The present system identifies semantic orientation of an opinionated phrase as either positive or negative. The classification of text as either subjective or objective is clearly a precursor to determine the opinion orientation of evaluative text since objective text is not evaluative by definition. A subjectivity classifier has been used to perform sentence level subjectivity classification. The present system is a hybrid approach to the overall opinion polarity identification problem and works with lexicon entities and linguistic syntactic features. The baselines system works only with SentiWordNet (Bengali) 2 . The use of lexical features like negative words, stemming cluster, functional word and parts of speech improved the performance of the present system over baseline. Inclusion of the chunk feature has improved the precision of the system by 19.2%. A further improvement of 3.6% in precision of the system has been obtained with the use of dependency relations information. Evaluation results of the final system have demonstrated a precision of 70.04% and a recall of 63.02%. KeywordsOpinion Mining, Polarity Identification, Bengali and Phrase Level Polarity Identification.